Definition of Large Number in Statistics

Definition of large number: large in what sense?

In ordinary conversation, a large number is simply a value that feels big. In statistics, that description is incomplete. A number might be large compared with a household budget but small compared with a national budget. Its size has meaning only after you identify what it measures and what you are comparing it with. Statistics has no universal numerical threshold above which every value becomes large.

The phrase can also refer to the number of observations in a study. That is a different idea: sample size describes how much information was collected, not how big the measured values are. A survey can record very large incomes from a small group of people, or modest incomes from a large group. To interpret either study, ask whether “large” describes a value, a count, or the strength of the evidence.

Large values are not large samples

Suppose someone reports a high total for visits to a community library. The total alone cannot tell you whether an average visit count is typical, whether visits are rising, or whether a survey represents all library users. Those questions require a denominator, a time period, and a description of how the observations were gathered. A total and a sample size can both be counts, but they play different roles.

A large sample means many observations relative to the task at hand. Even then, no single sample-size rule works for every question. Estimating a common preference with moderate precision is different from studying a rare event or comparing small subgroups. The variability of the outcome, the precision you need, and the sampling design all matter. OpenStax’s Introductory Statistics develops these distinctions through sampling distributions and estimation; the NIST/SEMATECH handbook likewise treats sample size in the context of a particular statistical procedure.

What the law of large numbers says

The law of large numbers gives “large” a more specific role. Under suitable conditions, such as repeated independent observations drawn from the same distribution with a finite mean, the sample average approaches the population mean as the number of observations grows. If the outcome is an indicator for whether an event happened, its sample average is the observed proportion. That proportion approaches the event’s probability under the same kind of conditions.

This is a statement about a long-run pattern, not a promise that a particular large sample must be close to the truth. An average can move away from the population mean when another observation is added. Nor does the law say that a short run is owed a balancing result: past outcomes do not make an independent event change its probability. OpenStax discusses the law as a reason that long-run averages stabilize, while its treatment of sampling distributions explains why estimates still vary from sample to sample.

A worked example with synthetic responses

Consider a library asking whether visitors would use a later closing time. Every count and result in this paragraph is synthetic and illustrative, not a report of an actual survey. In one synthetic random sample of 20 visitors, 14 say yes, giving a synthetic sample proportion of 14 ÷ 20 = 70%. In another synthetic random sample of 200 visitors, 116 say yes, giving a synthetic sample proportion of 116 ÷ 200 = 58%. The synthetic larger sample has a lower proportion; being larger does not force an estimate to rise, fall, or match a particular value.

If the library repeatedly took samples using the same sound method, estimates from larger samples would generally vary less from sample to sample than estimates from smaller samples. That is the relevant advantage, not a guarantee that the synthetic 58% is correct. We cannot determine the population proportion from these two synthetic samples alone. We would also need to know who could be selected, whether selection was random, and whether nonresponse might have changed the mix of people answering.

Imagine instead that the library asked only visitors who arrived shortly before closing. A much larger version of that survey could still misrepresent all visitors: the people easiest to reach might be especially interested in later hours. Collecting more answers from the same unrepresentative group does not repair that coverage problem. Sample size helps with random variation, but it cannot by itself remove systematic bias.

Why “big enough” depends on the question

There is a related but distinct idea called the central limit theorem. In common settings, it describes how the distribution of a sample mean can become approximately normal as sample size increases. The law of large numbers concerns convergence of the average toward a population value; the central limit theorem concerns the shape of its sampling distribution. Neither supplies a universal cutoff at which every dataset is suddenly reliable or approximately normal.

Whether a sample is big enough depends on the outcome’s distribution and what you intend to calculate. An uncommon outcome may need more observations before its estimated rate is useful. Strongly skewed measurements can make approximations less dependable at sample sizes that work well for less skewed measurements. Subgroup analysis also divides the available observations: a substantial overall sample can leave little information about a particular subgroup.

Start with the decision you need to make. Is the goal a rough description, a narrow interval estimate, or a comparison that could affect policy? Then examine the sampling method and the uncertainty appropriate to that goal. The NIST/SEMATECH handbook emphasizes checking the assumptions behind statistical methods rather than treating sample size as a substitute for those checks.

How to use the phrase clearly

When you encounter “a large number,” identify its unit first. Is it a large amount of money, a large count of events, or a large number of observations? Next, ask for the comparison that makes it large. A raw count can change simply because the population or observation period changed; a rate or proportion may provide a fairer comparison.

For a sample, ask what population it is meant to represent and how observations entered it. More data can make a well-designed estimate steadier, but it cannot make an unrepresentative selection representative. Precise wording keeps these claims separate: “large total” describes magnitude, “large sample” describes how many observations were collected, and “more precise estimate” is a conclusion that requires assumptions about the data.

Practical checklist:

• Name the quantity and its unit.
• State what “large” is being compared with.
• Check how the sample was selected before relying on its size.
• Report uncertainty when using a sample to estimate a population value.